RU2520424C2 - Method for complexion digital multispectral images of earth's surface - Google Patents

Abstract

FIELD: physics, computer engineering.

SUBSTANCE: invention relates to means of processing location images of the earth's surface. The method comprises determining the most informative image by calculating characteristic entropy of each image, calculating the morphological shape of the most informative image based on histogram segmentation with a given number of histogram modes, calculating the morphological projections of the rest of the images on the shape of the most informative image. Complexion is then carried out by summing pixel brightness of the most informative image which is taken as the basic image, and projections of the rest of the images on the shape of said image.

EFFECT: clearer scene objects on an image.

7 dwg

Description

The present invention relates to passive remote sensing and radar, in particular to the processing of location images of the earth's surface.

The prior art method of image formation is described in patent RU 2171499 (published July 27, 2001), which includes scanning the observed surface using several CCD arrays having mutual overlap in the field of view, obtaining several digital grayscale images, selecting a pair of images , one of which is accepted as the base, and the other is optional, the selection of objects of the same name in the images and combining the image by forming common areas.

The technical result is the formation of an enlarged field of view by high-precision seamless geometric and photometric combining of individual images obtained from various CCD arrays.

The disadvantage of this method is the need to highlight areas of the same name objects, which requires additional computational costs, as well as obtaining CCD image arrays in the same ranges of electromagnetic radiation.

The prototype of this invention is a method for combining digital multispectral grayscale images described in patent RU 2342701 (publ. 12/27/2008), which includes obtaining source images, decomposing each source image into low-frequency (LF) and high-frequency (HF) components, improving the characteristics of image components , which is an adaptive filtering and adaptive correction of brightness and contrast, separate processing of the low- and high-frequency components of images, the integration of image components, main Based on the principle of weighted summation for each pixel, the formation of the resulting image. The technical result is to obtain an image of high quality containing informative elements of images of the same scene obtained in different spectral ranges.

The technical result is achieved by the fact that each source image is subjected to multi-level decomposition by a Haar wavelet by means of a fast discrete stationary two-dimensional wavelet transform in order to obtain an approximating component, which is the LF component of the image, and a family of detail components, which are the RF components of the image.

The disadvantage of this method is the need to separate the source image from each channel into low-frequency and high-frequency components by decomposition of the Haar wavelet, which inevitably leads to information loss and distortion during sharp changes in the brightness of the image pixels in the form of steps of different brightness with a size of several pixels.

The technical effect, the solution of which the claimed invention is directed, is to obtain a more informative image containing elements of the original images of the same scene, obtained in different spectral ranges.

The indicated technical effect from the implementation of the proposed invention is achieved due to the fact that in the method for combining digital multispectral images of the earth’s surface, which includes obtaining source images, complexing the components based on the principle of weighted summation for each pixel, after obtaining the image, the most informative image is determined by calculating the intrinsic entropy of each image, the morphological form of the most informative image is calculated zheniya histogram-based segmentation of a predetermined number of modes of the histogram is calculated morphological projections rest on the shape of the most informative image of the image, and the image aggregation is performed using the most informative image, which is taken as the base and the projections rest on the shape image of the image.

The method is implemented by measuring the information content of the source image for the subsequent selection of the reference image, for this, calculate the intrinsic entropies of each image by the formula

$$E\left(X\right)=-{\displaystyle \sum _{k=\mathrm{one}}^K{w}_k\left(X\right)\cdot {\log }_2\left[w_k\left(X\right)\right]},\left(\mathrm{one}\right)$$

where w _k (X) is the one-dimensional probability density of the image X, which characterizes the distribution of pixel brightness over the image area.

The image, which is characterized by the highest value of E (X), is more informative.

Before calculating the morphological form, images need to determine auxiliary functions. The image model in a uniform field of view X is a piecewise constant function

$$f\left(x\right)={\displaystyle \sum _{i=\mathrm{one}}^N{c}_i\cdot {\chi }_i\left(x\right),x\in X.\left(2\right)}$$

Moreover, the field of view of X is divided into regions A _i ∈X, i = 1, 2, ..., N. All points A _i have the same brightness c _i ,

$${\chi }_i\left(x\right)=\{\begin{array}{l}\mathrm{one,}x\in A_i,\\ 0x\notin A_i,\end{array}\left(3\right)$$

.- indicator function of the set A _i , A _i ∩ A _j = ⌀ for i ≠ j, i = 1, 2, ..., N; $${\displaystyle \underset{i=\mathrm{one}}{\overset{N}{\cup }}{A}_i=X}$$

.

The accuracy of determining the boundaries of the regions of the partition depends on the number of levels of the partition N. Then dividing the image into sets of the same brightness can be called the image shape and calculate it as follows:

$$V_f=\left\{f\left(x\right)={\displaystyle \sum _{i=\mathrm{one}}^N{c}_i\cdot {\chi }_i\left(x\right),x\in X,c_i=\left(-\infty ,\infty \right)},i=\mathrm{1,2}...,N\right\}.\left(\mathrm{four}\right)$$

Next, we calculate the morphological projections of the remaining images on the form of the most informative image:

$$P_{V_f}g\left(x\right)={\displaystyle \sum _{i=\mathrm{one}}^N\frac{\left(g,{\chi }_i\right)}{{\Vert {\chi }_i\Vert }^2}{\chi }_i}.\left(5\right)$$

изображения g на форму V_f является изображением из множества V_f, наиболее близким к g, которая показывает различия по форме двух изображений f и g.Orthogonal projection $$P_{V_f}g$$

image g on the form V _f is the image from the set V _f closest to g, which shows the differences in the shape of the two images f and g.

Moreover, (g, χ _i ) in expression (5) is the scalar product of two functions with respect to the coordinates {x ₁ , ... x _K ):

$$\left(g,{\chi }_i\right)={\displaystyle \sum _{k=\mathrm{one}}^{K}g\left(x_k\right)\cdot \chi \left(x_k\right),\left(6\right)}$$

- норма индикаторной функции:a $$\Vert {\chi }_i\Vert$$

- norm indicator function:

$$\Vert {\chi }_i\Vert ={\left({\displaystyle \sum _{k=\mathrm{one}}^K{\chi }^2}\left(x_k\right)\right)}^{\raisebox{1ex}{$\mathrm{one}$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}.\left(7\right)$$

The distance between images f and g is determined by the norm of the difference f-g:

$$\rho \left(f,g\right)=\Vert f-g\Vert ={\left({\displaystyle \sum _{k=\mathrm{one}}^K{\left[f\left(x_k\right)-g\left(x_k\right)\right]}^2}\right)}^{\raisebox{1ex}{$\mathrm{one}$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}.\left(8\right)$$

для других каналов.The integration algorithm is based on the principle of weighted summation of the image pixels of each channel. The number of channels N must be at least two. The projection of the remaining images onto the shape of this image is added to the most informative image. Each additional term determines the contribution of one of the remaining channels to the resulting image. The weights for one channel are the distances $$\Vert f-P_{V_f}g\Vert$$

for other channels.

$$\begin{array}{l}{f}_c\left(x\right)=f_{\mathrm{one}}\left(x\right)+P_{f_{\mathrm{one}}}{f}_2\left(x\right)\cdot \Vert f_2\left(x\right)-P_{f_{\mathrm{one}}}{f}_2\left(x\right)\Vert +...\\ ...+P_{f_{\mathrm{one}}}{f}_N\left(x\right)\cdot \Vert f_N\left(x\right)-P_{f_{\mathrm{one}}}{f}_N\left(x\right)\Vert .\left(9\right)\end{array}$$

As a result of integration, a more informative image of high quality is formed.

The operation of the algorithm is illustrated by the structural diagram in the drawing (figure 1).

After obtaining the initial two-dimensional digital grayscale images (Img ₁ , Img ₂ , ..., Img _N ), the procedure for determining the most informative image 1 is carried out, then the morphological form V of the most informative image 2 is calculated, as well as the morphological projections of the remaining images are calculated (P _Img1 , P _Img2 , P _ImgN ) on the form of the most informative image Img _i 3, then the most informative image, as well as the morphological projections of the remaining images on the form of the most informative image are exposed I’m integrating procedure 4, based on the principle of weighted summation with weight coefficients, which are the distances between the base image and the morphological projections of the remaining images on the shape of the base image.

The technical effect achieved from the proposed invention is to increase the information content of images by combining measurement information from several spectral channels based on spectrozonal differences of objects in the image.

As an example, consider halftone images of the optical range of electromagnetic radiation in three spectral sub-bands - red (R), green (G) and blue (B), respectively. This algorithm was implemented using the MATLAB package.

The original images in three different ranges R, G and B are shown in figa, b, c.

The information content of the images is equal to E (R) = 7.66; E (G) = 7.35; E (B) = 6.12; as a result, the image of the R range is the most informative.

и $$f_B-P_{f_R}{f}_B$$

приведен на фиг.3а, б.In the calculations, the entire range of brightnesses was divided into several subranges A _i (A _i = 8), for which indicator functions, shapes and projections were calculated. The result of highlighting the differences in the scenes in form $$f_G-P_{f_R}{f}_G$$

and $$f_B-P_{f_R}{f}_B$$

shown in figa, b.

Integration is carried out according to the following formula:

. $$f_c\left(x\right)=f_R\left(x\right)+P_{f_R}{f}_G\left(x\right)\cdot \Vert f_G\left(x\right)-P_{f_K}{f}_G\left(x\right)\Vert +P_{f_K}{f}_B\left(x\right)\cdot \Vert f_B\left(x\right)-P_{f_R}{f}_B\left(x\right)\Vert$$

.

The synthesized image is shown in figure 4.

The resulting image has an information content of 7.73, which exceeds the information content of each of the original images and makes it possible to more clearly observe the objects of the scene during registration in different ranges of electromagnetic radiation.

The proposed method of complexing allows to increase the information content of images and more efficiently use remote sensing data for further processing and analysis.

Claims (1)

A method for combining digital multispectral images of the earth’s surface, including obtaining source images, component integration based on the principle of weighted summation for each pixel, characterized in that after obtaining the image, the most informative image is determined by calculating the intrinsic entropy of each image, the morphological form of the most informative image is calculated based on histogram segmentation with a given number of histogram modes s, calculate the morphological projections of the remaining images on the shape of the most informative image, and the integration is carried out by summing the brightness of the pixels of the most informative image, which is taken as the base, and the projections of the remaining images on the shape of this image.

Images